Chronic obstructive pulmonary disease (COPD) is the third leading cause of death globally. Characterized by a progressive worsening of lung function, it often goes undiagnosed until after the onset of severe symptoms.
Doroodgar Jorshery
"All of this shows the need for better approaches to identifying those patients at risk for developing COPD," said Saman Doroodgar Jorshery, MD, MPH, a postdoctoral fellow at the Massachusetts General Hospital Cardiovascular Imaging Research Center (MGH CIRC) in Boston.
One such approach involves the use of routine chest X-ray images (CXR).
"Over two billion CXRs are performed every year," Dr. Doroodgar Jorshery added. "Although these images may not be diagnostic for COPD, that's not to say they lack information."
The key to extracting that information is AI.
Leveraging a Widely Used Imaging Modality
During a Thursday session, Dr. Doroodgar Jorshery presented an AI tool called CXR-Lung-Risk that can predict a patient's long-term risk for developing COPD using nothing more than routine chest X-ray images.
MGH CIRC originally developed the CXR-Lung-Risk convolutional neural network as a tool for predicting lung-related mortality from CXRs. Having proved successful there, the team conducted a study to have it validated as a means of identifying individuals at high-risk of incident COPD over six years using routine outpatient CXRs.
Analyzing CXRs from 10,913 patients who have a regular history of smoking and 15,582 non-smoker patients, all with no history of lung cancer or COPD, the AI model proved capable of identifying individuals at high-risk for incident COPD beyond currently used clinical risk factors. According to the CXR-Lung-Risk model, individuals in the high-risk group had a five times higher risk of incident COPD among smokers and a three times higher risk among non-smokers.
"CXR is a widely used imaging modality, and the opportunistic screening of existing CXRs using a deep learning-based model could help identify high-risk individuals and guide COPD prevention," Dr. Doroodgar Jorshery said.
Although CXR-Lung-Risk stratified risk between individuals, prospective studies are needed to test whether the model can improve COPD prevention.
A Giant Leap Towards Personalized Medicine
According to Dr. Doroodgar Jorshery, this study adds to a growing body of evidence that AI can extract information from routine medical imaging to predict long-term risk of chronic disease. For example, in the case of COPD, the technology can augment existing clinical risk factors, leading to accurate, early diagnoses of the disease.
"The CXR-Lung-Risk tool is but one example of how AI can enhance a radiologist's ability to better predict long-term risk using routine imaging like chest X-rays," Dr. Doroodgar Jorshery concluded. "While this represents a big step for radiology, it marks a giant leap forward towards achieving personalized medicine for individual patients."
Access the presentation, "Leveraging Deep Learning of Chest Radiograph Images Identifies Individuals at High-Risk of Chronic Obstructive Pulmonary Disease," (R1-SSCH09-3) on demand at Meeting.RSNA.org.
